Light source fingerprinting using machine learning techniques

Abstract

Nowadays, Sensors that use light are used in a variety of applications because of their accuracy and convenience. For example, in newer electronic devices such as tablets and smartphones, it is used to detect light and adjust the brightness of the screen. And in automatic doors, it is also used to determine whether or not a person is in front of the door. However, sensors that detect light are vulnerable to attacks using the same kind of light source. One of the related works has demonstrated that a valid attack can be made using IR light sources in the medical infusion pump which uses a light-sensing sensor. Unfortunately, most sensors themselves cannot prevent such attacks. We propose research about classifying light sources (3 types of LED, 3 types of Laser) of the same kind using machine learning technology as part of how to effectively defend with such attacks on sensors that detect light. For the fingerprinting research of light sources, we created a stable light source data collection environment using spectrometers. Then, we measure the classification accuracy by applying machine learning classification models to the collected data to see if it was classifiable. We also analyzed which section of the light source's data is the most important as part of the interpretation of the machine learning results. Besides, we suggested additional research directions that can be developed from our study.